Browsing soundscapes

Abstract

Browsing soundscapes and sound databases generally relies on signal waveform representations, or on more or less informative textual metadata. The TM-chart representation is an efficient alternative designed to preview and compare soundscapes. However, its use is constrained and limited by the need for human annotation. In this paper, we describe a new approach to compute charts from sounds, that we call SamoCharts. SamoCharts are inspired by TM-charts, but can be computed without a human annotation. We present two methods for SamoChart computation. The first one is based on a segmentation of the signal from a set of predefined sound events. The second one is based on the confidence score of the detection algorithms. SamoCharts provide a compact and efficient representation of sounds and soundscapes, which can be used in different kinds of applications. We describe two application cases based on field recording corpora.